Page 1
Department of Economics and Finance
Working Paper No. 1916
http://www.brunel.ac.uk/economics
Eco
nom
ics
and F
inance
Work
ing P
aper
Series
Zara Canbary and Charles Grant
The Marginal Propensity to Consume for Different Socio-economic Groups
October 2019
Page 2
The Marginal Propensity to Consume
for Different Socio-economic Groups
Zara Canbary∗ and Charles Grant†
∗zara.canbary(at)brunel.ac.uk†charles.grant(at)brunel.ac.uk
1
Page 3
Abstract
This paper investigates the marginal propensity to consume for the UK
households across different socio-economic groups. It uses the Family Expendi-
ture Survey, a repeated cross section of British Households, which reports expen-
diture, income, and household characteristics from quarter 1 of 1986 to quarter 1
of 2016. Since each household is interviewed only once we construct pseudo-
panels based on the socio-economic status of the household head. We find that
households with higher socio-economic status have lower marginal propensity to
consume. We also find that the marginal propensity to consume increased after
the 2007-2009 financial crisis. This study supports the hypothesis that credit
constraints are more serious for lower income groups.
Keywords:
JEL classification: D1, D9, D14.
Page 4
1 Introduction
Estimates of the marginal propensity to consume from changes in income have usually
found that households are more sensitive to changes in income than is predicted by
the Permanent Income Hypothesis. Hall (1978), for example, argued that 20 percent
of households are “rule-of-thumb” consumers and spend a fixed proportion of their
current income. A common explanation for this excess sensitivity to changes in cur-
rent income is that some households are liquidity constrained (Flavin 1984). Such
households are unable to smooth consumption since they are unable to borrow in pe-
riods where their income is below the desired level of consumption implied by the
Permanent Income Hypothesis. A key problem is to identify households likely to be
credit constrained. Hayashi (1985), for example, argues households with low levels
of savings are constrained, while Zeldes (1989) argues households with low assets-to-
income ratios are constrained. They both find their constrained households are more
sensitive to income changes and that around 15-20 percent of households do not follow
the Permanent Income Hypothesis.
In this paper, rather than use the level of savings or assets as a proxy for credit
constraints, we will argue that there are differences in access to credit across socio-
economic groups. Professional households are not only likely to have higher and more
stable income than low-skilled households, but they are also more likely to have ac-
cess to credit. This paper will use British household data for 1986-2016 to estimate
the marginal propensity to consume and compare the response to anticipated income
1
Page 5
changes for four different socio-economic groups. The key contribution of the paper
is that it is the first paper that compares the marginal propensity to consume of dif-
ferent types of British household. It will examine how the four socio-economic groups
differ and whether these differences are consistent with the hypothesis that lower socio-
economic groups are more likely to be liquidity constrained, and hence more sensitive
to changes in their income.
An important advantage of our data-set is that it includes the period before, during
and after the 2007-2009 financial crisis; the period associated with major changes in
borrowing criteria which restricted access to credit markets, (See: Bank of England
Financial Stability Report, Sep 2008). This enables us to study the effect of the finan-
cial crisis on household expenditure. In this paper we will explore how the marginal
propensity to consume of the four different socio-economic groups differ before and af-
ter the financial crisis. Our hypothesis is that the crisis affected lower socio-economic
groups more severely than those households in higher socio-economic groups.
The paper proceeds as follows: Section 2 discusses the existing literature on the
Permanent Income Hypothesis in more detail. Section 3 gives a detailed description of
household data which is used in this study. Section 4 describes the empirical method-
ology while section 5 reports the results. The conclusions are described in section 6.
2
Page 6
2 Literature Review
A large literature has been published on the marginal propensity to consume with many
showing how household consumption responds to changes in economic resources. The
Permanent Income Hypothesis, as outlined by Milton Friedman (1957), suggests only
permanent and unexpected income shocks result in a major revision in consumption.
This theory suggests that people use borrowing and saving to smooth income fluctu-
ations and they should not respond to changes in income that are fully anticipated.
Therefore, an estimation of the marginal propensity to consume out of anticipated
income changes should yield insignificant results. For example, an anticipated promo-
tion at work, that can result in change in income level, should not affect the marginal
propensity to consume at the time it happens since the expectation of the income
change is already included in the information set. Instinctively, when lagged consump-
tion and income are included as instruments in regression a consumption decision is
made based on information available at time t− 1. Hence, the marginal propensity to
consume out of predictable changes in income on the basis of past information should
be statistically insignificant.
The theory also suggests that rational agents’ desired consumption is determined
by permanent income, while they have access to credit market; suggesting that when
households face a temporary reduction in income to continue consuming as before they
need to have access to debt to finance this consumption. This is important because,
for example, if a group of households are excluded from the credit market, they are
3
Page 7
likely to react strongly to anticipated changes in income.
The permanent income hypothesis has been tested and rejected over time with
liquidity constraints as one of the main reason for rejecting the hypothesis. Hall et
al. (1978) demonstrates that given the inclusion of lagged consumption, no other
variable observed in earlier periods should have any explanatory power for current
consumption. He finds households respond differently to different sources of income
variations and concludes that aggregate consumption should be modelled for optimal
choice of a single, fully rational, and forward looking agent ie. Euler equation approach.
Hall (1978) rejects the implications of the pure life cycle-permanent income; arguing
households display “excess sensitivity” to predictable changes in income. His results
suggest 80 percent of households follow the permanent income hypothesis, but that 20
percent of households are “rule-of-thumb” consumers who consume a fixed proportion
of their current income. Hall (1978) does not mention the reason for rejection of
hypothesis.
Similarly Flavin (1985) tests the Permanent Income Hypothesis using US Annual
Aggregate data and shows marginal propensity to consume to be different from zero
and reports excess sensitivity for the proportion of the population subject to liquidity
constraints. This could not be attributed to myopic behavior of the individual since
the inclusion of unemployment rate as the proxy for liquidity constraint changes the
marginal propensity to consume both in magnitude and significance. Without the
liquidity constraint proxy, she finds the marginal propensity to consume to be 0.37.
After inclusion of the unemployment rate as part of the information set, the marginal
4
Page 8
propensity to consume falls to 0.15; significantly different from the initial estimate. She
states that a lack of access to credit market and the myopic behavior of individuals
are the main reasons. Both these papers reject the version of the Permanent Income
Hypothesis with perfect capital markets. In both papers, predictable changes in income
are shown to affect changes in current consumption.
Hayashi (1985) also argues Permanent Income Hypothesis applies to about 85%
of the population and income changes explains only a small fraction of the movement
in expenditure. He also shows households with high levels of savings are associated
with lower excess sensitivity. Later, Altonji and Siow (1987) using PSID data finds
including the coefficient of lagged income growth, that the marginal propensity to
consume out of predicted changes in income is statistically significant. Zeldes (1989)
households’ asset to income ratio as measurement of liquidity constraints to confirm
the excess sensitivity. He concludes that households with higher asset to income ratio
were consistently less sensitive to income changes. Poterba (1988), Wilcox (1989),
and Campbell and Mankiw (1989) present analysis of reactions to predictable changes
in income using aggregate data. They show that periods in which consumption is
high relative to income are typically followed by rapid growth in income. They find a
significant marginal propensity to consume of between 0.32 and 0.71. Their findings
suggest that while most most households seem to follow the simple rule-of-thumb
model of consumption, for a fraction of forward-looking households, their knowledge
of future income growth is reflected in current consumption and hence they follow the
Permanent Income Hypothesis.
5
Page 9
The relationship between liquidity constraint and consumption, in the light of per-
manent income hypothesis, has received considerable attention from economists. It is
worthwhile to look at some studies that consider evidence from individual households
expenditure surveys. Runkle (1991) considered home-ownership status as measure
of ease of access to borrowing. He assumes that home-owners are less constrained
and show less excess sensitivity. He directly tests for liquidity constraints using panel
data on individual households and finds no evidence of liquidity constraints. He sug-
gests that the failure of the permanent-income hypothesis is due to aggregation bias.
Jappelli et al (1998), exploited the Survey of Consumer Finance to estimate the prob-
ability of a household being constrained. They studied food consumption changes in
response to anticipated income changes from Panel Study of Income Dynamics and
found no evidence for much excess sensitivity associated with the possibility of con-
straints. Later, Jappelli et al. (2010) established the probability a household was
denied access to credit and refused Permanent Income Hypothesis for households with
lower probability of access to credit.
Shapiro and Slemrod (1995) interviewing households after announcement of tax
reduction concluded that 40% of people interviewed planned to spend the extra cash.
Taking the predictable nature of this transitory income increase, Souleles (2002) ex-
ploited the anticipated income increase induced by pre-announced tax refunds to test
the Permanent Income Hypothesis. Given the predictable nature of this changes in
income, it should thus not alter consumption in the year of its receipt, he finds that
consumption is excessively sensitive to anticipated tax-cuts with a marginal propen-
6
Page 10
sity to consume of 35% to 60%. In a similar paper, Parker (1999), using the CEX,
studied the reaction of household consumption to predictable changes in social secu-
rity taxes using the security payroll cap, a predictable income decrease in January and
increase in the middle of the year. The results show 20 cents increase in non-durable
consumption for each dollar increase in this anticipated income. He also rejected the
possibility of households being liquidity constrained since the sample only included
high-income taxpayers. Similarly, Shapiro and Slemrod (2009) used 2008’s tax rebate
as a case of predictable income increase and showed that this mostly led to an increase
in expenditure for 20% of survey respondents.
There are few studies that support the excess sensitivity for the households. For
example, Browning and Collado (2001) is using ECPF Spanish panel data and insti-
tutionalized June and December extra wage payments to full-time workers as a case
of anticipated income increase and finds no evidence of excess sensitivity suggesting
bounded rationality as a reason why earlier researchers found large response of expen-
diture to predicted income changes. Hsieh (2003) used both annual payments from
the state of Alaska’s Permanent Fund and tax rebates as cases of predictable income
increase and only finds evidence for excess sensitivity with respect to tax refunds but
not with respect to payments from the state of Alaska’s Permanent Fund.
The literature we have reviewed has largely rejected the Permanent Income Hy-
pothesis since changes in consumption are excessively sensitive to predictable changes
in income. One major criticism of this literature is that many of papers are using na-
tional aggregate data. Attanasio and Weber (1993) argue that such data is subject to
7
Page 11
aggregation biases, and more importantly conceals the heterogeneity in consumption
behaviour across different types of household. They advocate using household survey
data where the income changes are traced for each family over time. Using such data
allows us to capture the consumption behaviour of households with different household
characteristics. However, there are limited number of household panel data sets avail-
able with relatively small sample size that often experience attrition and non-response.
Hence, most existing studies have been conducted on using US data. The data used
in our study is not a true panel. Instead, following Browning, Deaton and Irish (1985)
and Attanasio and Weber (1993), we construct a Pseudo-panel. We group individuals
who share the same socio-economic status into cohorts, and use the averages within
these cohorts as observations in our pseudo panel.
A further criticism of this literature is the nature of proxy for the credit constraint.
For example, McCarthy (1995) and Jappelli used level of wealth, Pistaferri (2012)
cash-on-hand, Zeldes (1989) used asset to income ratio, and Runkle (1991) used home-
ownership to classify the households as constrained or unconstrained. These commonly
used factors suggest a very narrow view of credit market conditions.
A number of papers have explored how consumption changed during the 2007-2009
financial crisis. For example, Jensen et al (2017) using Danish household data show
banks that reduced their lending caused a significant decrease in the borrowing and
spending of their customers. They also find that borrowing remained lower after the
crisis and spending foregone during the crisis has not recovered. Dutt et al (2009),
using US data, find similar evidence when businesses are unable to borrow during and
8
Page 12
after the financial crisis. However, to the best of our knowledge, there is no paper
which investigate the changes in behavior of UK households; and in particular how
the marginal propensity to consume of households changed as a result of the crisis.
The aim is to see if the estimate of the marginal propensity to consume out of
predictable changes in income varies across different socio-economic groups and if so,
to see how they are affected, in both magnitude and statistical significance, during
and after the financial crisis.
3 Data
This paper uses UK household data on consumption and income from 1986 Q 1 to
2016 Q 1. The Family Expenditure Survey (FES), compiled by the Office for National
Statistics, has detailed information on the income and spending of a large number of
individual households, covering mainland Britain and Northern Ireland, but excluding
students in residential halls, the armed forces, people living in nursing and residential
homes, prisoners and the homeless. As well as detailed responses to questions on in-
come and expenditure, the survey also reports details on household characteristics such
as age, household size, household composition, and socio-economic status. However, it
does not include any information on households’ level of education. Each wave of the
FES reports the responses of around 6000 households. Households are interviewed on
a continuous basis throughout the year, although each household is only interviewed
once (meaning the survey is not a genuine panel).
9
Page 13
The FES was discontinued in 1994. Between 1995 and 2002 it was replaced by the
Expenditure and Food Survey (EFS). Although it categorized the expenditure vari-
ables in a slightly different way, the main change is that the survey replaced paper
questionnaires with directly digitally recorded responses. It is nevertheless comparable
with the earlier FES survey. This survey was renamed the Living Costs and Food Sur-
vey (LCF) in 2002 when changes were made to make it comparable to other household
surveys in the rest of the European Union. This last change resulted in the some slight
changes in the individual expenditure categories.
The use of FES is prompted by Attanasio and Weber (1995). They encourage the
use of micro household data rather than the aggregate data commonly used in the
study of household consumption and argue that the individuality of agents are better
preserved in Survey data, hence, more useful when studying households’ behaviour.
Additionally, we combine data from the FES, the EFS and the LCF surveys. Thus it
will use data from 1986 to the first quarter of 2016. The data was combined using the
2001 consumption categories contained in the Living Cost and Food Survey (known
as Classification of Individual Consumption by Purpose, COICOP). This allows us
to construct a harmonized overall measure of total and non-durable consumption for
each household that is constructed consistently between the surveys. Combining the
surveys using identical definitions of the consumption categories enables us to have
thirty years of data, a considerably longer period than each individual survey covers.
The questions on income are the same across all three surveys. There are separate
questions on wages, second jobs, self-employed income, non-wage income and social
10
Page 14
transfers (e.g. benefits). The key question we exploit in our analysis is the question
‘what is the normal weekly disposable income of the household?’. This formulation of
the question has some advantages. While it is not necessarily the household’s income in
any particular week, it will be a measure of the household’s normal (or expected) level
of income, and thus, we claim, a good proxy for the households’ permanent income.
It is changes in permanent income (or normal income) which should cause changes
in the level of consumption of the household (according to the Permanent Income
Hypothesis), rather than unpredictable and temporary changes in current income.
The survey data used in this paper is compared to the National Account data
in figure 1. The figure shows the average level of overall consumption in the three
household surveys (using the left-hand scale and plotted with a solid line), and average
household consumption given by the national account data (using the right-hand scale
and plotted with a dashed line).1 The household survey data uses three different
surveys, and the figure shows that there is a break in 1992 when the survey switched
from the FES to the EFS, and a further break when the survey switched from the
EFS to the LCF. Nevertheless, average household consumption grew steadily through
most of the sample period. The data shows there was a small recession in the early
1990s and a small decline in 2007 (the height of the recession which resulted from
the sub-prime crisis). The pattern of consumption in the three household surveys is
1Note that the numbers are not completely comparable since the National Account data will
include household spending by care-home residents, prisoners, military members, and tourists but
excludes holiday spending. It will also include the spending made by unincorporated businesses.
11
Page 15
similar that shown in the national account data. The major difference seems to be the
sub-prime recession was longer and deeper in the national account data than in the
LCF. Nevertheless, the similarities in the broad trends gives us confidence that the
use of the survey data is sensible.
3.1 Constructing Pseudo-Panel
Since households are only interviewed once in the household surveys, we can not con-
struct a true household panel. This problem can be overcome by following the approach
suggested in Browning, Deaton and Irish (1985); creating a pseudo-panel with the use
of cohorts from repeated cross-sections where we create groups of households with
shared characteristics. In this approach, individuals sharing some common charac-
teristic are grouped into cohorts, and the average level of consumption and income
within each time period and for each cohort is constructed. Both Deaton (1985) and
Attanasio and Weber (1995) used year-of-birth to define the cohorts, while Maki et al
(2001) defined cohorts based on the level of education.
The key issue we investigate in our paper is the marginal rate of consumption
for different groups. We will define groups which are likely to differ in the extent
to which they are liquidity constrained. Kempson and Whyley (1999), looking at
US data, argued that employment status and ethnicity were good determinants for
whether a household is excluded from borrowing. Demirguc-Kunt and Klapper (2013)
found that age and employment status are also good predictors of whether a household
has access to credit markets. Unfortunately the households do not report their level
12
Page 16
of education in each of the waves of the survey used in this study. Hence in this
study we will define the cohorts based on the socio-economic group of each household.
We construct four socio-economic groups, “Professional”, “Skilled”, “Unskilled”, and
“Unoccupied”; households with a higher socio-economics status are less likely to be
liquidity constrained, and hence socio-economic groups are a good proxy for the level
of financial exclusion the household experiences.
While the pseudo-panel is not a true panel, since the same households are not used
in both time periods, it nevertheless does have some advantages. The key advantage
is that the sample response rate will not change over time, since, unlike a true panel,
it will not suffer from attrition. As a result, the results from using a pseudo-panel may
well be more reliable.
We then investigate the relationship between expenditure and income. Other im-
portant factors determining consumption including real interest rates, household char-
acteristics such as age of the household reference person, number of adults plus number
of children to make up the family size are also included in the consumption function
as control variables.
Table 1 reports summary statistics of household disposable income and expenditure
by socio-economic cohort. Household expenditure in each category of consumption as
percentage of disposable income is presented in parentheses for each socio-economic
groups. The average weekly disposable income is shown in column 3. It is at the highest
for the “Professional” households at £905.00 and the lowest for the “Unoccupied”
households at £310.40. Weekly average total expenditure of households follows the
13
Page 17
same trend. It is reported in column 4 and it is the highest in value at £696.00 which
is about 77% of disposable income for “Professional” households. It decreases to
£567.00 for “Skilled” households however, at 92% there is an increase as percentage of
disposable income for “Skilled” households compared to the “Professional” households.
The average weekly expenditure decreases again for “Unskilled” households to £471.00,
however, as the percentage of their disposable income, there is an increase to 97.5%
compared to the “Skilled” households. Weekly average total expenditure is the lowest
at £314.00 for the “Unoccupied” households. This socio-statistic group has the highest
expenditure level as the percentage of their disposable income compared to other
groups at 105.7%.
This trend persists for the expenditure on non-durable goods that is reported in
column 5 of table 1. Expenditure on non-durable goods and services consists about
54.5% of households total expenditure out of disposable income. It is £489.50, 54.5% of
their disposable income, for“Professional” huseholds. There is an increase in spending
on non-durable goods and services as percentage of disposable income as the household
socio-economic status moves from higher to lower skilled employment. Expenditure on
non-durable goods and services is £410.00, 66.5% of disposable income, for “Skilled”
households. It is £351.80, 73.50% of disposable income, for “Unskilled” households
and It is £237.45, 79.70% of disposable income, for “Unoccupied” households. This
table shows that households in higher socio-economic groups consume lower percent-
age of their disposable income in each category of expenditure compared to those in
lower socio-economic groups. This is specially important results, since by design, the
14
Page 18
households available funds after their normal average expenditure on goods and ser-
vices are deducted, determines the amount of credit entitlement for the households.
Therefore, as the percentage of expenditure out of the disposable income increases,
the amount of credit a household can access decreases.
3.2 Financial Crisis
We believe that the 2007-2009 recession is likely to have had an important effect on the
behavior of households. Access to saving and borrowing is a necessary for households to
smooth their consumption. The ability of households to obtain credit was dramatically
affected by the policy changes after the financial crisis. The Credit Conditions Survey
by Bank of England2 reports a fall in the availability of secured and unsecured credit to
households since mid-September 2008 with a view to further reduction in the coming
months, Bank of England (2008). This financial crisis transmitted into real economy
in October 2008 when the Bank of England started lowering the interest rate initially,
from 5% to 4.5%, and eventually falling to 0.5% in March 2009.
Table 2 shows the time line of events happened between 2007-2009 that resulted
in one of the worst global financial crisis in history. The initial warning signs came
early in 2007, when three major US mortgage providers folded during the sub-prime
2Credit Conditions Survey is a quarterly survey released by Bank of England in which Lenders
are asked about secured and unsecured lending to households, to non-financial corporations, small
businesses, and to non-bank financial firms in the past three months and the coming three months.
The survey is used by the Bank of England’s to assess the latest developments in bank funding and
household and corporate credit conditions.
15
Page 19
mortgage crisis. The crisis later spread across Europe, including UK, causing volatility
in the stock market. The UK government had to bail out faltering banks, including
temporary nationalisation of the Northern Rock. The crisis deepened in the summer
of 2008 when Lehman Brothers, after being refused a bailout by the US government,
announced their bankruptcy. This incident caused panic amongst global bankers,
leading to the Great Recession. The stock market crashed shortly afterwards. Banks
become reluctant to loan and credit markets continued to tighten. Figure 2 shows how
consumer credit fell sharply in 2007-2008. This slow down in credit hits the lowest in
2008.
It was thought that easy lending and mortgage default are a key reason for behind
the financial crisis, as well as the changes in interest rate. We have divided the sample
into two periods, where the break point is at the end of the third quarter in 2008 as
banks increasingly tightened their lending criteria. This follows Blinder (2013) who
define the beginning of the credit crunch to be the bankruptcy of Lehman Brothers.
This is the point at which the access to credit was harder and limited resulting in a
reduction in credit to the household sector. This reduction in credit is likely to have
affected the ability of households to smooth consumption; in particular, an ability
to borrow during and after the financial crisis is expected to affect the capacity of
households to manage temporary income declines.
We explore the effect of the financial crisis on household consumption. The aim
is to find out if households’ marginal propensity to consume differs before and after
the financial crisis to see whether the crisis resulted in a change in the households’
16
Page 20
marginal propensity to consume. We also investigate whether this change was larger
for households in lower socio-economic groups compared to those in higher socio-
economic groups. We expect a relatively low marginal propensity to consume for
households in higher socio-economic groups, who are likely to be able to maintain the
credit access before and after the financial crisis. In contrast, we expect the marginal
propensity to consume for lower socio-economic households, who are more likely to
be credit constrained, both have a higher marginal propensity to consume before the
crisis, and to increase their marginal propensity to consume after the financial crisis
when the borrowing becomes more difficult.
4 Methodology
An aim of this study is to look at the marginal propensity to consume for UK house-
holds and investigate if it differs for households in different socio-economic cohorts.
Analysing the data with a simple model we propose testing the hypothesis under
Permanent Income Hypothesis. Within the permanent income hypothesis, marginal
propensity to consume out of anticipated changes in income should be close to zero.
If the hypothesis is rejected, consumption displays excess sensitivity.
Similar to much of the previous work on the Permanent Income Hypothesis, we
estimated the augmented version of the Euler Equation. We consider four different
socio-economic groups and some control variables for household characteristics.
17
Page 21
∆ lnCit = α +4∑
i=1
βi∆ lnYit + γrt + λZit + εit (1)
On the left hand side, we have change in logarithm of the consumption, ∆ lnC for
group i between periods t− 1 and t.3 On the right had side, we have the predictors of
changes in consumption growth; the measure of predictable income changes, ∆ lnY ,
and the real interest rate, rt and control variables for the household characteristics Z,
ε is the error term. The subscript i denotes the socio-economic groups. These cohorts
are defined for “socio-economic” status of the households; “Professional”, “Skilled”,
“Unskilled”, and“Unoccupied”. The regression includes the interest rate rt and a set
of controls for household characteristics, Z. We follow Pistaferri (2001), and include
time-varying components such as family size and age.
The equation 1 is estimated for total consumption and consumption of non-durable
goods and services. The key variable of interest is β, indicating the marginal propensity
to consume out of anticipated changes in income. The implications of the permanent
income hypothesis we expect β ∼= 0. This in turn implies that changes in consumption
are not predictable, thus delivering the well known martingale consumption result
(Hall, 1978). Previous income is certainly one of most important determinants of
household consumption and needs to be controlled in order to properly evaluate income
change on consumption level. To overcome this problem we use the instrumental
variable method of estimation to generate an unbiased estimation of β.
3Following Jensen’s inequality, the arithmetic average of logarithm of reported values are calculated
for expenditure and income rather than the customary logarithm of the arithmetic average.
18
Page 22
In the first stage anticipated changes in income is regressed on the instruments to
obtain coefficients that reflect the amount of variation in income changes attributable
to this set of instruments.
The first stage coefficient is used to generate predicted value for income changes. This
predicted income contains all the information set held by agents up to time t− 1 that
helps them make expenditure decisions. This predicted value of income changes is
used to obtain an estimate of the relation of expenditure behaviour and changes in
past values of income changes.
For the implementation of the GMM approach, following Blinder and Deaton
(1985), Flavin (1981), and Hall (1978), we use as our instrument four lags of income
changes. An innovation in this paper is that we also use consumer confidence indicator,
lagged once, as an instrument in addition to four lags of income changes. We are using
the consumer confidence index constructed by GfK Consumer Confidence Index; a sur-
vey designed to capture individuals’ attitudes regarding the current and perceived near
future economy status, it is affected by economic news, uncertainty, economic growth,
and current economic situation amongst many other economic factors. The Consumer
Confidence Indicator measures how confident people feel about their income’s stabil-
ity. Hence, it impacts households’ economic decisions such as spending activity. As
a result, consumer confidence is a key indicator for the overall shape of the economy.
The inclusion of lagged Consumer Confidence Index as a forward looking variable is
to capture the effect that is not included in economic fundamentals. Individual agents
form rational expectations for future income subject to the individuals’ information
19
Page 23
set at time t−1, Ωt−1. Examples of such information could be possibility of promotion
at work or financial literacy of the agents that are hard to capture adequately from
our data set.
We tested the power and validity of the instruments; four lags of income changes
and consumer confidence indicator, lagged once. The values of the F statistics is 35.93.
The power of the instruments easily exceeds the conventional minimum standard of
power of F = 10. In addition, Hansen’s (1982) test for over-identification is consistent
with the validity of our instruments. The J-statistic follows a chi-square distribution
with 4 degrees of freedom. We fail to reject the null hypothesis that the instruments
are valid.
The variables used in equation 1 are expected to capture the variation in the
marginal propensity to consume for households in different socio-economic groups.
As well as reporting results for the full sample, we also reports results for two
sub-periods; before and after the financial crisis of 2007. This enables us to investi-
gate whether the marginal propensity to consume changed during the financial crisis.
We anticipate that the financial condition of household, borrowing and credit access,
changed during the financial crisis due to the changes in bank’s lending policies. If
households access to credit changed then it will affect their marginal propensity to
consume after the crisis. Our data includes the Financial Crisis of 2007 during which
a change in borrowing criteria and tightening of the financial conditions limited house-
holds’ credit access significantly.4 These changes were communicated with the public
4See: Financial Stability Report by the Bank of England, October 2008.
20
Page 24
prior to implementation allowing the households to adjust consumption a few quarters
before it actually occurs. As explained in section 2, we consider September 2008 as
the point of expected tightening of the borrowing conditions. We then evaluate the
marginal propensity to consume out of a households expected change sin income before
and after the crisis.
5 Results
In this section, we report the results for the marginal propensity to consume for differ-
ent socio-economic groups. We defined as consumption all expenditure items except
mortgage and rent payments. The expenditure values are inflation adjusted to the
2015 price level. The regression equation 1 included income growth instrumented by
four lags of changes in income growth and lag of changes in the consumer confidence
index and it is augmented by controls for a set of household characteristics including
family size, age, and the real interest rate. We established the marginal propensity
to consume from predicted changes in income using the full sample, and two sub-
samples, before and after the financial crisis of 2007. We ran separate regressions for
total expenditure and the expenditure on non-durable goods and services.
Results for the marginal propensity to consume out of the expected changes in
income are reported in table 3. Results are shown for the full sample of households in
columns 1-2, as well as the sub-samples from before the financial crisis in columns 3-
4, and the sub-samples after the financial crisis in columns 5-6. The results suggest
21
Page 25
households have different levels of marginal propensity to consume depending on the
households’ socio-economic status.
Results for the change in total consumption for the full sample is reported in
column 1. Results show that the marginal propensity to consume (MPC) out of
expected changes in income is not statistically significant for the professional (socio-
economic group 1) and the skilled (socio-economic group 2). The MPC is is 0.94
for unskilled households (socio-economic group 3), and statistically significant at 1%.
The MPC for unoccupied (socio-economic group 4) is 0.75 and significant at 5%. The
results also indicate an increase in the MPC out of expected changes in income for
total household consumption as the socio-economic status increases; apart from the
unusually high coefficient for the unskilled (socio-economic group 3). The existing
literature, (See: Flavin, 1984 and Campbell et al, 1989), reports the MPC between
0.3 and 0.7. While our results for the professional and skilled households (socio-
economic groups 1 and 2) at 0.53 and 0.59 are similar to the existing literature, the
MPC seems to be much higher for unskilled and unoccupied households at 0.94 and
0.75 respectively.
The second column in table 3 reports the marginal propensity to consume (MPC)
of non-durable goods and services out of expected changes in income for households in
different socio-economic groups. Results are significant for all four categorise of house-
holds with the lowest MPC of 0.47 for the professional Households (socio-economic
group 1). With the exception of the skilled households (socio-economic group 2) with
MPC of 0.93, MPC gradually increases to 0.65 for unskilled and 0.80 for unoccupied
22
Page 26
households. Coefficients are statistically significantly different from each other.
These results indicates that households with different socio-economic firstly alter
their expenditure when permanent income changes, secondly the degree at which they
alter the expenditure is different in households with different socio-economic groups.
These results are consistent with our belief that socio-economic status is a good proxy
for households access to credit.
5.1 Financial Crisis
To explore the effect of the financial crisis, we divided our data into two sub-samples;
the period up to 2008, and the period after 2008. If the financial crisis reduced the avail-
ability of credit to the household sector, then we would expect the marginal propensity
to consume from predicted changes in income to increase after the crisis. Moreover,
it is likely that the change is not the same for households in different socio-economic
groups.
Results for changes in total consumption prior to the financial crisis is reported
in column 3, in table 3. The marginal propensity to consume prior to the financial
crisis of 2007 follows the same trend and magnitude of those for the full sample.
The marginal propensity to consume out of changes in permanent income on non-
durable goods and services are reported in column 4. There are insignificant differences
between the coefficients for full sample and the sub-sample prior to the financial crisis
of 2007. Coefficients are different from each other for household in different categories
of households and households in socio-economic group 1 have the lowest marginal
23
Page 27
propensity to consume compared to those in higher socio-economic groups. socio -
economic level of household is a good proxy for households’ access to credit.
Estimates of the marginal propensity to consume for total consumption after the
financial crisis is reported in column 5. With the exception of the professional house-
holds in socio-economic group 1, the trend of increasing magnitude persists. However,
the coefficients are different from those prior to the financial crisis shown in column 3,
both in magnitude and statistical significance. It is interesting results for professional
households in socio-economic group 1 since the marginal propensity to consume has
changed from 0.4 and statistically insignificant prior to the financial crisis to significant
at 1% after the financial crisis of 2007. However, the coefficient is not significantly
different from 1. The coefficient for households in group 2 and group 3 are not sta-
tistically significant. However, surprisingly, the marginal propensity to consume out
of predicted changes in income on non-durable goods and services for unskilled house-
holds in socio-economic group 3 show significant decrease after the financial crisis. It is
0.54 and not statistically significant after the financial crisis compared to the marginal
propensity to consume for the same group of households prior to the financial crisis of
2007 that was 0.93 and significant at 5%.
Marginal propensity to consume(MPC) out of predicted changes in income for non-
durable goods and services after the financial crisis of 2007 is reported in Column 6,
table 3. With the exception of unskilled households in socio-economic group 3, the
marginal propensity to consume for non-durables follows the same trend as the total
consumption. The marginal propensity to consume for the the professional households
24
Page 28
in socio-economic group 1 has significantly increased to to proximately one after the
financial crisis of 2007, same result is indicated for unoccupied households in socio-
economic group 4. The results do not show the same increase across the households
from higher to lower groups, however, the marginal propensity to consume is statisti-
cally different from each other for households in different socio-economic groups.
6 Conclusion
A basic assumption of the permanent income hypothesis is that individuals have free
access to the credit market, lending and borrowing at the same rate enabling house-
holds to smooth consumption as the current income level changes. According to the
permanent income hypothesis predictable changes in permanent income should not
alter consumption level; the coefficient, β, should be approximately zero.
Overall results show that for most part professional and skilled households, indicate
lower marginal propensity to consume compared to unskilled and unoccupied. While
Hall (1978) states around 80% of the households plan their expenditure following the
permanent income hypothesis, our results suggest this percentage to be around 50%,
when investigating total expenditure. Results for the full sample expenditure on non-
durable goods and services also rejects the permanent income hypothesis, indicating
the marginal propensity to consume of more than zero and significant for all four
socio-economic groups. Our findings are consistent with those of Flavin (1993), who
is using unemployment as a proxy for liquidity constraints.
25
Page 29
The financial crisis had a significant effect on the households’ expenditure be-
haviour. Prior to the 2007 financial crisis, the results suggest that around half of
households were following the permanent income hypothesis. However, the results af-
ter financial crisis shows only professional households that is a quarter of households
follow the permanent income hypothesis.
Results for the consumption of non-durable goods and services is even more in-
teresting as it indicates the same drop in the percentage of households following the
permanent income hypothesis when setting their expenditure level. The marginal
propensity to consume for the households in socio-economic groups 1-2 is mostly in
line with the permanent income hypothesis. After the crisis, it is unskilled household
in socio-economic group 3 that are still following the permanent income hypothesis.
This study aims to contribute to this growing area of research by exploring marginal
propensity to consume for different socio-economic groups in the United Kingdom.
Initially, we have estimated the marginal propensity to consume using UK Household
Expenditure Survey and found evidence that the household expenditure pattern was
significantly affected by the anticipated changes in income. These results are consistent
with a significant proportion of households being credit constrained.
We have defined four different socio-economic groups, and considered the status as
proxy for credit market access: households with higher socio-economic groups are likely
to be less liquidity constrained. The Permanent Income Hypothesis was then tested
for the four different groups and the response to predictable changes in income rejects
the hypothesis for skilled, semi-skilled and unoccupied households. Results support
26
Page 30
the idea that professional households, who are least likely to be credit-constrained, are
more likely to follow Permanent Income Hypothesis. It also confirms that households
with lower socio-economic status have higher marginal propensity to consume. The
evidence presented by this paper when using household data adds further support
to earlier studies in the rejection of the Permanent Income Hypothesis. Firstly, our
findings show that households react to anticipated changes in income by altering their
consumption. Secondly, and more interestingly, the marginal propensity to consume
out of anticipated changes in income is significantly lower for households in upper
socio-economic status. This gradually falls when moving from upper socio- economic
groups to lower socio- economic groups.
This alteration in consumption is even more significant during and after a financial
crisis, with tightening of credit by banks as one possible explanation. This resulted
in an increase in marginal propensity to consume, with the poorer household showing
a higher increase in marginal propensity to consume in the aftermath of the financial
crisis.
The second finding is of more significant consequences since it confirms that house-
holds that are excluded from credit market by design, are unable to smooth consump-
tion when income changes.
27
Page 31
References
[1] Aghion, P. and Bolton, P. (1997). A Theory of Trickle-Down Growth and Development.
The Review of Economic Studies, 64(2), p.151.
[2] Altonji, J. and Siow, A. (1987). Testing the Response of Consumption to Income
Changes with (Noisy) Panel Data. The Quarterly Journal of Economics, 102(2), p.293.
[3] Attanasio, O. and Weber, G. (1995). Is Consumption Growth Consistent with Intertem-
poral Optimization? Evidence from the Consumer Expenditure Survey. Journal of Po-
litical Economy, 103(6), pp.1121-1157.
[4] Banerjee, A. and Newman, A. (1993). Occupational Choice and the Process of Devel-
opment. Journal of Political Economy, 101(2), pp.274-298.
[5] Blinder, Alan S. and Angus Deaton (1985), ”The Time Series Consumption Function
Revisited”, Brookings Papers on Economic Activity, no. 2, pp. 465-521.
[6] Blundell, R., Pistaferri, L. and Preston, I. (2008). Consumption Inequality and Partial
Insurance. American Economic Review, 98(5), pp.1887-1921.
[7] Browning, M. and Collado, M. (2001). The Response of Expenditures to Anticipated
Income Changes: Panel Data Estimates. American Economic Review, 91(3), pp.681-
692.
[8] Browning, M, and Lusardi, A. Household Saving: Micro Theories and Mi-
cro Facts. Journal of Economic Literature 34, no. 4 (1996): 1797-855. http://
www.jstor.org/stable/2729595.
28
Page 32
[9] Bucks, B. and Pence, K. (2008). Do borrowers know their mortgage terms?. Journal of
Urban Economics, 64(2), pp.218-233.
[10] Campbell, J. and Mankiw, N. (1989). Consumption, Income, and Interest Rates: Rein-
terpreting the Time Series Evidence. NBER Macroeconomics Annual, 4, p.185.
[11] Carbo,S. and Gardener,E. and Molyneux, P. (2007). Financial exclusion in Europe.
Public Money and Management, 27 (1) , p. 21
[12] Carroll, C., Slacalek, J., Tokuoka, K. and White, M. (2017). The distribution of wealth
and the marginal propensity to consume. Quantitative Economics, 8(3), pp.977-1020.
[13] Deaton, A. (n.d.). Life-cycle models of consumption: is the evidence consistent with the
theory?. [S.l.]:[s.n.].
[14] Demirguc-Kunt, A. and Klapper, L. (2013). Measuring Financial Inclusion: Explaining
Variation in Use of Financial Services across and within Countries. Brookings Papers
on Economic Activity, 2013(1), pp.279-340.
[15] Devlin, J.F.(2005). A detailed study of financial exclusion in the UK Journal of Con-
sumer Policy, 28 , pp. 75-108
[16] Flavin, M. (1985). Excess Sensitivity of Consumption to Current Income: Liquidity
Constraints or Myopia?. The Canadian Journal of Economics, 18(1), p.117.
[17] Flavin, M. (1993). The Excess Smoothness of Consumption: Identification and Inter-
pretation. The Review of Economic Studies, 60(3), p.651.
[18] Friedman, M., 1957. A Theory of the Consumption Function, Princeton NJ: Princeton
University Press.
29
Page 33
[19] Galor,O. and Zeira, J. (1993) Income Distribution and Macroeconomics, The Review
of Economic Studies, Volume 60, Issue 1, Pages 35-52
[20] Hall, R. and Mishkin, F. (1982). The Sensitivity of Consumption to Transitory Income:
Estimates from Panel Data on Households. Econometrica, 50(2), p.461.
[21] Hayashi, F. (1985). The Permanent Income Hypothesis and Consumption Durability:
Analysis Based on Japanese Panel Data. The Quarterly Journal of Economics, 100(4),
p.1083.
[22] J.M. Hogarth, C.E. Anguelov, J. Lee Who has a bank account? Exploring changes over
time, 1989-2001 Journal of Family and Economic Issues, 26 (1) (2005), p. 7
[23] Hsieh, C. (2003). Do Consumers React to Anticipated Income Changes? Evidence from
the Alaska Permanent Fund. American Economic Review, 93(1), pp.397-405.
[24] Jappelli, T. and Pistaferri, L. (2010). The Consumption Response to Income Changes.
SSRN Electronic Journal.
[25] Jensen, T. and Johannesen, N. (2017). The Consumption Effects of the 2007-2008 Fi-
nancial Crisis: Evidence from Households in Denmark. American Economic Review,
107(11), pp.3386-3414.
[26] Kempson E., Whyley C.(1999). Kept out or opted out? Understanding and combating
financial exclusion Policy Press, London
[27] Parker, J. (1999). The Reaction of Household Consumption to Predictable Changes in
Social Security Taxes. American Economic Review, 89(4), pp.959-973.
30
Page 34
[28] Poterba, James M. “re Consumers Forward Looking? Evidence from Fiscal Experi-
ments.” The American Economic Review, vol. 78, no. 2, 1988, pp. 413-418.
[29] Pistaferri, L. (2001). Superior Information, Income Shocks, and the Permanent Income
Hypothesis. Review of Economics and Statistics, 83(3), pp.465-476.
[30] Runkle, D. (1991). Liquidity constraints and the permanent-income hypothesis. Journal
of Monetary Economics, 27(1), pp.73-98.
[31] Shea, J. (1994). Should we test the Life Cycle: Permanent Income Hypothesis with food
consumption data?. Economics Letters, 45(1), pp.63-68.
[32] Shea, J. (1995). Union Contracts and the Life-Cycle/Permanent-Income Hypothesis.
The American Economic Review, Vol. 85, No. 1 (Mar., 1995), pp. 186-200.
[33] Stephens, M. (2003). “3rd of tha Month”: Do Social Security Recipients Smooth Con-
sumption Between Checks?. American Economic Review, 93(1), pp.406-422.
[34] Shapiro, M. and Slemrod, J. (2009). Did the 2008 Tax Rebates Stimulate Spending?.
American Economic Review, 99(2), pp.374-379.
[35] Slemrod, J. and Shapiro, M. (1995). Consumer Response to the Timing of Income.
Cambridge, Mass.: National Bureau of Economic Research.
[36] Simpson, W. and Buckland, J. (2009). Examining evidence of financial and credit exclu-
sion in Canada from 1999 to 2005. The Journal of Socio-Economics, 38(6), pp.966-976.
[37] Souleles, N. (1999). The Response of Household Consumption to Income Tax Refunds.
American Economic Review, 89(4), pp.947-958.
31
Page 35
[38] Yuan, Y. and Xu, L. (2015). Are poor able to access the informal credit market? Evi-
dence from rural households in China. China Economic Review, 33, pp.232-246.
[39] Zeldes, S. (1989). Optimal Consumption with Stochastic Income: Deviations from Cer-
tainty Equivalence. The Quarterly Journal of Economics, 104(2), p.275.
32
Page 36
Figure 1: Households’ Average Total Expenditure in the FES and National Accounts
FES
EFS
LCFS
600
800
1000
1200
NA
Ann
ual H
ouse
hold
Exp
endi
ture
100
200
300
400
500
FES
Ann
ual H
ouse
hold
Exp
endi
ture
1980 1990 2000 2010 2020
wave
Notes: This figure plots the households’ average expenditure in the Household Survey and National
Accounts. The continuous black line representing the Household Survey data is our own calculation
using UK household expenditure survey data from first quarter 1996 to first quarter 2016 for survey
based line. The dashed gray line represents the National Accounts is from ONS for National Accounts
data. The left axis is the households’ average annual expenditure calculated using the Family Expen-
diture Survey. The right axis is the households’ average annual expenditure from National Account
data.
33
Page 37
Figure 2: UK Consumer Credit
Notes: Changes of total (excluding the Student Loans Company) sterling gross consumer credit
lending to individuals (in sterling millions) seasonally adjusted. Source: Bank Of England.
34
Page 38
Table 1: Summary Statistics, tab:tab1, updated 04/06/19
Socio-economic Observations Disposable Total Non-durablesGroup Income Consumption ConsumptionProfessional 40,107 905.00 696.00 489.50
(77%) (54%)Skilled 34,378 590.00 567.00 410.00
(96%) (69%)Unskilled 37,879 524.50 471.00 351.80
(90%) (67%)Unoccupied 31,218 310.43 314.00 237.45
(101%) (76%)
Notes: Source: Own calculation using UK household expenditure survey data from
first quarter 1996 to first quarter 2016. All values are in British Pounds. Prices are
deflated using the BOE price index for year 2015 to convert nominal prices to current
prices. Expenditure as percentage of disposable income in parentheses.
35
Page 39
Tab
le2:
Eco
nom
ican
dF
inan
cial
Eve
nts
2007
-200
9
2007
Jan
Su
b-p
rim
ed
own
grad
e(U
S)
Ow
nit
Mort
gage
Solu
tions
Inc.,
Am
eri
can
Fre
edom
Mort
gage,
Inc.,
Mort
gage
Lenders
Netw
ork
USA
Inc.
fold
ed.
Feb
HS
BC
war
nin
g(U
S)
The
bad
debt
pro
vis
ions
for
2006
tob
e20%
hig
her
than
exp
ecte
dto
roughly
$10.5
bn.
Mar
Su
b-p
rim
eco
llap
seSub-p
rim
ele
nders
decla
red
bankru
ptc
y,
inclu
din
gA
ccre
dit
ed
Hom
eL
enders
Hld
g,
New
Centu
ryF
in.,
DR
Hort
on
&C
ountr
yw
ide
Fin
.
Aug
UK
stock
mar
ket
vola
tili
ty(U
K)
Banks
begin
tost
op
lendin
gto
each
oth
er
due
tom
ark
et
fears
.
Sep
Nor
ther
nR
ock
cris
is(U
K)
Nort
hern
Rock
sought
em
erg
ency
fundin
gfr
om
the
BO
E,
firs
tru
non
abank
for
more
than
acentu
ry.
2008
Feb
Nat
ion
aliz
edN
orth
ern
Rock
(UK
)Str
uggling
Nort
hern
Rock
isto
be
nati
onalized
for
ate
mp
ora
ryp
eri
od.
Mar
Bea
rS
tear
ns
Bai
lou
t(U
S)
The
firm
was
bought
out
by
JP
Morg
an
Sep
Leh
man
Bro
thers
Ban
kru
ptc
y(U
S)
Govern
ment
pro
tecti
on
for
Lehm
an’s
$60
bn
inuncert
ain
mort
gage
ass
ets
was
reje
cte
d.
Nat
ion
oliz
edF
orti
s(E
U)
Euro
pean
bankin
gand
insu
rance
gia
nt
was
part
lynati
onalized
toensu
reit
ssu
rviv
al.
Bra
dfo
rdB
ingl
ey(U
K)
The
govern
ment
takes
contr
ol
its£50
bn
of
mort
gages
and
loans.
Savin
gs
op
era
tions
and
bra
nches
are
sold
toSpain
’sSanta
nder.
HB
OS
cris
is(U
K)
Ithad
tob
ere
scued
by
Llo
yds
TSB
aft
er
ahuge
dro
pin
its
share
pri
ce.
Was
hin
gton
Mu
tual
and
Wac
hov
ia(U
S)
Tw
om
ore
Am
eri
can
banks
collapse
d.
Iris
hb
ankin
gcr
isis
The
firs
tE
uro
pean
countr
yto
dri
ftin
tore
cess
ion,
pro
mis
efr
om
Gov
tounderw
rite
the
enti
reIr
ish
bankin
gsy
stem
,ult
imate
lyunable
to
uphold
.
AIG
Nat
ion
aliz
ed(U
S)
Abailout
of
$150
bn
toin
sure
loans
and
mort
gages
again
stdefa
ult
.
Cre
dit
Mar
kets
Fro
zeB
anks
hoard
ed
cash
($160
bn)
tow
rite
-dow
nbad
mort
gages
and
wit
hdra
wals
that
led
toa
cash
short
age;
resu
ltin
gin
Zero
inte
rest
rate
.
Oct
Low
erin
gIn
tere
stR
ate
(UK
)T
he
Bank
of
Engla
nd
cuts
inte
rest
rate
by
0.5
%to
4.5
%to
ease
the
financia
lcri
sis.
Tro
ub
led
Ass
etR
elie
fP
rogr
am(U
S)
$700
Bn
Bailout
bill
was
set
up
topurc
hase
toxic
mort
gages
from
banks.
RB
S,
TB
S,
and
HB
OS
Bai
lou
t(U
K)
The
thre
ebanks
receiv
ed
Govern
ment
resc
ue
package
of£37
bn.
Nov
En
dof
Inve
stm
ent
Ban
kin
g(U
S)
Gold
man
Sachs
and
Morg
an
Sta
nle
yb
ecam
ere
gula
rcom
merc
ial
banks.
End
of
an
era
of
dere
gula
tion
and
hig
hri
sk.
Sto
ckM
ark
et
Cra
shM
ark
et
plu
mm
ete
ddue
toC
ongre
ssre
jecti
on
of
$60
bn
the
bank
bailout
bill.
Dec
Mor
ego
vern
men
tb
ailo
uts
(US)
Cit
igro
up
and
the
auto
indust
ryals
ore
ceiv
ed
bailouts
.
2009
Jan
Blu
eM
ond
ayC
rash
(UK
)B
riti
shbankin
gsh
are
scollapse
d.
Ban
kof
Am
eric
aB
ailo
ut
(US
)P
rovid
ed
wit
h$20bn
from
$700
bn
financia
lre
scue
fund
tohelp
itw
ith
the
loss
es
incurr
ed
when
itb
ought
Merr
ill
Lynch.
Feb
RB
Sfo
lds
(UK
)R
BS
rep
ort
sa
loss
of£24.1
bn
for
2008,
the
big
gest
inB
riti
shcorp
ora
tehis
tory
,m
ade
togiv
eup
an
annual
pensi
on
wort
hab
out£700,0
00.
Notes
:S
ourc
e:O
wn
calc
ula
tion
.In
this
table
we
pre
sent
ati
mel
ine
of
the
key
even
tsin
the
worl
dfi
nan
cial
cris
isb
etw
een
2007-2
009.
Th
em
ost
inte
nse
ph
ase
ofth
ecr
edit
cris
isw
asin
Sep
tem
ber
2008
,w
hen
the
ma
jor
US
inve
stm
ent
ban
kL
ehm
an
Bro
ther
sfi
led
for
ban
kru
ptc
y.
36
Page 40
Tab
le3:
Reg
ress
ion
Res
ult
sF
ull
Sam
ple
Yea
r<=
2008
Q3
Yea
r>20
08Q
3V
AR
IAB
LE
S∆c it
∆cn
dit
∆c it
∆cn
dit
∆c it
∆cn
dit
∆lnY
1t
0.53
70.
466*
0.40
10.
348
1.89
2***
1.45
4**
(0.3
45)
(0.2
70)
(0.3
80)
(0.2
77)
(0.5
81)
(0.6
30)
∆lnY
2t
0.59
40.
975*
**0.
638
0.95
2*0.
261
0.76
2*(0
.570
)(0
.369
)(0
.764
)(0
.516
)(0
.627
)(0
.414
)∆
lnY
3t
0.94
2***
0.65
1**
0.93
1**
0.72
9**
0.54
10.
269
(0.3
28)
(0.2
66)
(0.3
69)
(0.3
30)
(0.5
20)
(0.3
81)
∆lnY
4t
0.75
7**
0.80
1***
0.74
6*0.
705*
*0.
837*
*1.
045*
**(0
.312
)(0
.211
)(0
.421
)(0
.291
)(0
.427
)(0
.259
)r t
0.20
20.
0544
0.24
0-0
.054
1-0
.357
-0.2
83(0
.177
)(0
.124
)(0
.352
)(0
.206
)(0
.604
)(0
.472
)se
10.
0049
4-0
.030
5-0
.001
78-0
.044
60.
0979
0.06
69(0
.031
5)(0
.023
1)(0
.037
9)(0
.027
2)(0
.064
8)(0
.048
7)se
20.
0021
4-0
.032
4-0
.004
55-0
.048
3*0.
0850
0.06
06(0
.031
1)(0
.022
0)(0
.038
6)(0
.026
5)(0
.059
2)(0
.042
7)se
3-0
.014
9-0
.021
9*-0
.017
3-0
.029
2**
-0.0
0784
-0.0
0551
(0.0
162)
(0.0
118)
(0.0
207)
(0.0
148)
(0.0
271)
(0.0
209)
Av.
Age
-0.0
925
-0.1
35**
*-0
.119
-0.1
76**
*0.
0134
0.00
129
(0.0
687)
(0.0
503)
(0.0
881)
(0.0
652)
(0.1
19)
(0.0
806)
Av.
Age
Sq.
0.01
620.
0212
***
0.02
030.
0275
***
0.00
389
0.00
398
(0.0
107)
(0.0
0785
)(0
.014
1)(0
.010
5)(0
.018
2)(0
.012
5)A
v.
Fam
ily
Siz
e0.
0258
0.06
83**
0.03
930.
0934
**-0
.082
7-0
.054
3(0
.042
0)(0
.030
6)(0
.051
7)(0
.037
4)(0
.080
5)(0
.055
0)
Obse
rvat
ions
424
424
304
304
120
120
Tes
tof
over
id.
res.
(P-v
alue)
(0.6
)(0
.53)
(0.8
6)(0
.70)
(0.3
3)(0
.52)
Notes:
Inth
ista
ble
resu
lts
are
rep
ort
edfo
rp
oole
dre
gre
ssio
nan
dfu
llsa
mp
le.
Th
eL
HS
vari
ab
leis
the
gro
wth
into
tal
con
sum
pti
on
∆c i
tan
dth
egro
wth
inn
on
-du
rab
le
con
sum
pti
on
∆cn
dit
.T
he
vari
ab
le∆
lnYit
,i
=1−
4re
pre
senti
ng
fou
rso
cio-e
con
om
icgro
up
s,re
pre
sents
the
change
inlo
gari
thm
of
inco
me
at
tim
et.
∆ln
Yt
isin
stru
men
ted
wit
h
fou
rla
gs
of
inco
me,
an
dla
gged
chan
ges
inco
nfi
den
cein
dex
dCCI t
−1.
se1-s
e3are
du
mm
yvari
ab
les
for
diff
eren
tso
cio-e
con
om
icgro
up
s.r t
isth
eB
OE
real
inte
rest
rate
.R
ob
ust
stan
dard
erro
rsin
pare
nth
eses
.In
the
tab
le***
mea
ns
sign
ifica
nt
at
1p
erce
nt,
**
mea
ns
sign
ifica
nt
at
5p
erce
nt,
*m
ean
ssi
gn
ifica
nt
at
10
per
cent.
Ran
kT
est:
F(5
,424)=
35.
37